LAUSR

Deep learning has become a recent, modern technique for big data processing, with promising results and large potential. For recommender systems, user and item information can be used as input vectors to perform prediction tasks. However, augmenting the number of layers to improve feature extraction will increase the computational complexity considerably and may not achieve the desired results. This paper proposes a method called attention convolution collaborative filtering (Att-ConvCF), which integrates an attention mechanism with a collaborative filtering model to improve the effectiveness of the feature extraction by reassigning the weights of feature vectors. Descriptive documents for the items are used to enrich the background information through a convolutional neural network. Finally, extensive experiments with real-world datasets were performed, and the results showed that Att-ConvCF could effectively extract the feature values of the data and significantly outperform the existing recommendation models.